Economists normally calculate TFP as the residual of a regression on output against K, L, and maybe other variables.
However, a well behaved regression is expected residuals to be normal (https://en.wikipedia.org/wiki/Normality_test). This is at odds with a situation where TFP increases over time, as in real economies.
Is then the method of residuals flawed? What are some adjustments made to overcome this problem? Maybe adding a trend?
TFP is not simply a residual value in a regression as you have assumed, it is called residual because it is observed rather than measured. That is what probably causes your confusion.
It is what is left after the Labour and Capital inputs have been accounted for. Normally you can break TFP down further into components of technological progress, scale effect and technical efficiency.
There is plenty of literature and sometimes conflicting views on the measurement of TFP and growth accounting on the web. Perhaps the below links can serve as a starting point.
